Saving Democracy - With TV AI Realtime Fact-checking

Could Politicians be Real-time Fact checked on TV?

Could we see their track record Claim/Debunked numbers next to their name?

Think about it this way.

They mostly repeat the same debunked stories over and over.

But only machines could bring up numbers as counterarguments pretty much in real-time. It is essentially just a fuzzy search in existing fact-checking records.

You can't react a day later when poison reached the minds of millions.

You need to react in real-time or you will one day wake up in a much different world.

Imagine decentralized fact-checking where facts are mined/verified/rewarded the same way as any other crypto and use this undeletable blockchain of verified claims for live tv fact-checking.

Corruption and Lies will not be as easy as before ...

But let’s face it…

Humans are emotionally impulsive and easily manipulated creatures.

Millions of years old “US vs Them” gene from single cell organisms

Sure.

As you can see in the animation below. Pretty much every large imperium over a certain size sometimes internationally behaves like an elephant in a porcelain shop. Be it the recent US, Russia, UK, or China endeavors.

It pretty much always boils down to one man, typically seizing way too much power from the people, brainwashing all of them typically into doing something stupid in the name of “Us vs Them” even though the country boundaries are clearly virtual constructs.

As you can clearly see, there is literally no difference between humans and the primitive old genetically encoded assimilate or annihilate behavior they got since they were single-cell organisms, especially if you zoom out from the current time and space frame and look at

How overall borders moved during whole human history. by Ollie Bye See more at his fantastic Drex’s Youtube channel here.

It is the same story over and over of people in power, still sadly being driven by millions of years old primitive parts of the brain no matter what year it is.

But the US has something else going on too.

Something very special in this millions of years old Selfish-Gene dominated world of ours. Something that hopefully one day nationalistic fragmented Europa and the rest of the World will hopefully be able to emulate.

The US and The American Dream (while immigration was possible for people before my generation) in my opinion was the most beautiful multi-ethnic human experiment since Babylon.

A clear statement to how massively you can grow and what you can then achieve in Space, Science, or Medicine with talent from all over the world, if you are not limited by genes, race, or nationality and allow this now humongous human mass just pull together in one direction. In today’s nationalistic and typically primitive old genes-driven puristic world it was a rare experiment Indeed …

Thing is.

All world economies and democracies are currently completely interlinked with the US dollar and US political stability itself.

Sadly. A lot has radically changed since US Constitution was drafted.

Information no longer travels on slow horses or pigeons hitting just small towns at worst.

We now live in a world where thanks to technology one can now manipulate the minds of hundreds of millions in seconds without accountability over and over.

And because we are more hectic than ever we often just react not think. An explosive combination easily misused in automated way with AI by select a few with money.

World.

Where foreign governments can keep day by day fomenting and inciting lies and hate to activate very old gene or party purity bios routine in the human brain, and by violence and conflicts that ensue destabilize states from inside. Simply because it is untraceable and cheaper than conventional war. Why fight your enemies when you can make them fight each other.

Where as little as 3 people can shape and poison opinions of an entire country or even continent in a targeted and automated way.

State-sponsored Division sewing Hate factory,  2 GSM gateways, 3,000 SIM cards 18,000 fake accounts, press release.

Looking at this picture… Does Cambridge Analytica mass spreading targeted hate using AI to a precisely targeted group of US voters ring a bell?

So how many similar misinformation campaigns are in full swing on social media as we speak until the 2024 elections? The campaigns that are now able thx to AI and GPT clones generate messages personally to each radical group or person to stir maximum hate.

But how do you prevent stirring super-strong country destabilizing grade emotions with Lies?

Lies spread even by governments and Mainstream media like the Iraq war.

That media spread Lie in the end did cost 2 Trillion dollars and at least a million people dead.

Lie like Brexit where Lie traveling on the side of bus claiming that the UK sends £350 million a week to the EU pretty much sunk UK economy and EU’s military and economic strength.

In this modern technology age. Humans are clearly too slow to fact-check billions of messages in 10ms before the poison is delivered, the division is sown, and irreparable damage is done.

Maybe the only way to stop AI accelerated misinformation and hate spread to millions in 10 ms

IS

also superfast and scalable AI powered detection and factchecking also happening before message is shown and damage is done.

Perhaps with real-time fact-checking in any social media post (not censuring but marking it red short note explaining why and link to an interactive hierarchical list of educational Wikipedia with trusted academic and impartial sources with along proven track record claiming otherwise) impact of this can be reduced greatly

So now.

Believe it or not.

Nearly half of 360 million country now believe the Big Lie according to the latest Axios-Momentive pool from Jan 2022. Steep jump from a quarter of the country in the past Ipsos pool from May 21, 2021.

How could this even happen?

Just look at Youtube. Clickbait is now a major Ad selling currency.

We now live in a world where being radical is more important than being right.

Things like this are essentially bound to happen if the most-watched cable media keeps spewing hate and division just for a bigger view count and profit. Since radical opinions sell ads on tv more right?.

If you repeat Lie long enough thru 3-4 megaphones over and over it will become true for many that don't have time, will, energy or trust in mainstream media to verify said claims.

This toxic never-ending infight between two divided Americas keeps reminding me of an ancient but very relevant story. Since as we know. History frequently repeats itself if people just react and do not think.

The dying king and his three sons fighting each other for the throne.

They were fighting each other relentlessly for the right to the throne but the country was exhausted and weak because of this.

So he called them in and gave each bunch of tree branches and asked them to break them. They obviously couldn't.

Then he gave each just one branch and asked again. They broke it easily.

A lesson the US seems to have forgotten.

At least two US states now have not exactly small seccession movements. Some of which are tainted by way too much smoke from foreign adversaries.

January 6

Seeing when misinformation was being disseminated and attack on capitol and senate certifying election results of the oldest continuous democracy was incited and encouraged by a sitting president on live TV in front of our faces.

We all wished that someone would intervene.

But nobody did.

Why?

1. What can you do when an attack on Democracy is directed from the top of the power hierarchy with the biggest and loudest megaphone. Clearly, the constitution needs to take a lesson from nature and evolve to keep up with the rapidly changing technology world and new challenges that didn't exist when it was conceived. And Impeachment as a defense from kings is now pretty clear, will never work when half of the senate will happily rather give up democracy than lose the shrinking power. Most checks and balances had yes man installed probably even the supreme court creating an autocracy. In the end, judges and courts were actually the only check and balance left that just barely saved the democracy and will of the people. And to the so-called originalists, I say. There is no such thing as final software or life form in an ever-changing world.

   
   

2. Everything was normalized. Every moral norm eroded. a bit by bit. inaction by inaction year by year.

   
   

3. After decades of serving and reporting daily only about corporations and the rich elite. Mainstream media completely lost connection with and trust by 99% of common Americans due to serving and discussing only what the top 1% pay them for. Does CNN and MSNBC care or report daily about struggles and daily needs of rustbelt or middle class? Forget it. They are as spoiled and disconnected from on-the-ground reality as the TV studio in Hunger Games was. Fact happily exploited by the biggest radical con artists and worst of fringe outlets, now happily filling this new news source trust vacuum. We all remember CNN rather covering empty podium while ignoring massively popular rallies of candidates with anti-corporate stances or even passing debate questions to chosen candidates upfront. Or beating pro-IRAQ war and Terror Level drums for years. Trust is gone.

   
   

4. But the main reason half of the country is now believing a Lie is a fact that enforcing journalism standards is non-existent or behind paywalls. Even today. As we speak. The most viewed news networks happily keep spreading Lies and Hate by either creating or rebroadcasting them while having only one goal. View count and ad money essentially destabilizing a large portion of the country for a few bucks. A clearly missing legislative for journalism standard and inefficient threshold for license suspension rules.

   
   

5. Misinformation is intentionally packed with hints and half-truths keeping our defenses paralyzed. Preventing any reaction because there is always some truth in that half Lie, right?

   
   

   
   

But how can transparency of Journalism protect Democracy when access to Truth is restricted to the wealthy while Lies are Free to access for everyone?

Heck Lies will even find you.

Shouldn't easier access to Truth than to Lies prevent now half of the country to believe Big Lie?

Reality check. How many rust belt subscribers do they have behind paywalls?

Paywalls just created another vacuum happily immediately filled by Facebook conspiracy peddlers.

In a sense, NYT and WP are partly responsible for the current situation by giving up on free access to independent journalism for everyone.

Access to Truth should be right, not paid service exactly like Healthcare and Education.

If google search can live from ads without a paywall so should by able any independent journalists.

Paywalls are just lazy at best and definitely creating more echo chambers at worst. I mean wouldn't free access to Independent and fair Journalism be accessible to all and especially rustbelt states make sense?

Isn’t Free and easy access to Truth solution to currently very wobbly Legs of our Democracy?

Isn’t a free but Larger readership leading to larger ad revenue anyway? Google is not exactly a poor company.

But How did it all become so obscenely normalized so fast?

I feel like when the now-infamous bigoted and racist sentence regarding immigrant Mexicans first sounded on the Election candidate's rally on Live TV unchallenged.

“They are murderers and rapists and some are I suppose good people.”

Emboldened Charlottesville “Unite the Right” Rally (cc-by 2.0) followed soon by normalizing ...

“Good people on both sides”

Or. When a foreign adversary was asked to attack US infrastructure on Live TV watched by millions? And that attack was indeed immediately carried out as we later found?

Russia, if you’re listening, I hope you’re able to find the 30,000 emails that are missing

No microphone was cut. No fine was issued. No rally stopped. When clearly spreading and normalizing hate against an individual group or race or for calling for an adversary to attack the security infrastructure of your own country.

When you do nothing in situations like this, the following President John F. Kennedy's quote immediately comes to mind.

The only thing necessary for the triumph of evil is for good men to do nothing.

Sadly. The racist and checks and balances dismantling part were highly reminiscent of what was happening a not that long time ago in Germany.

Attack on the Reichstag building, home of the German parliament in Berlin, on Monday 27 February 1933, precisely four weeks after Adolf Hitler was sworn in as Chancellor of Germany (NAID) 535790)

Perhaps those who can't learn from history are bound to repeat it.

As for social media. The situation was and still is even worse.

In 10ms when Hate and Lies of one are spread to hundreds of millions or more.

Even if you delete a tweet day later and suspend an account. Most twitter message is being read by the time they are sent not a day later so damage is already done.

Cambridge Analytica used big data and AI to maximize misinformation and targeted hate speech efficiency in an automated way.

You cant put AI genie back to the bottle. AI mass misinformation attacks are a weapon and sad reality from any recent voting in US France and Britain.

And while we debate if using AI for Realtime factchecking politicians is fair to them, next ten Cambridge Analytica's are already guaranteed to attack next elections.

Photo: January 6. Attack on Capitol incited organized and normalized on Live TV. Derivative work: J JMesserly cc-by 2.0

Yet on the opposing receiving side.

Live TV or net mass AI hate and violence incitement detection is still virtually nonexistent or not even debated. Not because technology is not there.

But because legislative is not.

Banning people believing and further spreading misinformation is not a solution. Factchecking them live and accountability by showing red misstatement count next to name is.

Everyone who did learn misinformation surely has the capacity to learn better if incentivized.

How?

Most people that believe in big stolen election Lie don't watch mainstream media so anything published there will have zero effect.

But

If there is something like Mass Disinformation Prevention Act, essentially Law requiring

Mandatory total misstatement count one made during his public life being shown next to the name in order to get right to address crowds larger than 10000 if misstatement count bigger then 10 per year.

Even the blindest followers will be curious and surely will go check those numbers online. And that is the ultimate goal.

Numbers next to name will incentivize an actual act of verifying them, and establish a long missing True accountability of any Public Influencer

Even heavy propaganda can’t put a positive spin on it when the misstatement number is extremely high.

Mindset will be immediately changed from

blindly following

to curious

30000 lies shown next to name while others don 1 or 2 is hard to not notice.

But just silencing is not reducing the astonishing current number of confused souls.

And as we had seen. For tens of millions, the new communication platform will happily jump in in a week as Parlor did. In the end,

Parlor was denied hosting. But make no mistake new decentralized internet is coming where hosting can’t be easily denied anymore on Internet Computer (ICP).

Silencing people is also a double-edged sword.

Corporations should not be judges. The public should be.

This dangerous idea will be misused fast, depending on which side wins elections. Make no illusions. Democrats historically suppressed voting by redistricting via gerrymandering, just as Republicans did.

Live red subtitles with explanation and education generated on live TV or below the social post the moment the post is generated and misinformation detected.

Not 3 days later. Not 3 months later on a website that nobody visits.

But if you are planning to mark something as misinformation. Just claiming that statement is not true from a position of power without explorable independent facts is not enough.

You should provide essentially something like an infinite interactive Wikipedia-like experience with an ability to drill down to as many independently verifiable sources with an as good and long track record as possible.

For now, the link to factcheck.org provided drill-down should be better than nothing and still be fantastic added media stabilizing value.

This is the only way mainstream media can regain at least some trust and attention of general public and diminish growing influence of blatant misinformation sources poisoning minds daily

That is all nice.

But in reality. You have like 10ms. Billions of posts every second and humans are slow non-scalable.

Luckily current neural network language models like GPT3 are instant and infinitely scalable.

I intentionally didn't slap misused AI term to it since GPT3 is essentially a glorified SQL server. A new perhaps a bit fuzzy way to store retrieve and transform information while not being limited to just a subset of human language like SQL was.

And while you wouldn't call DB server necessarily an AI I will continue to use the term to not be rocking the hype boat too much.

Photo by imgix on Unsplash

So say hello to real-time AI fact-checking.

I proposed this Idea to OpenAI and since then have beta API access to the best language model AI we currently have - GPT 3

Now. Imagine you had good live voice recognition during for example any mass rally. Translating live spoken text to simple text that is easily processible by computer

Currently. If you prepare gpt3 with two-shot examples

Q: All black people are lazy

A: This is inciting hatred against a group or individual

and then feed Realtime each sentence from voice recognition of Politician's speech at his rally

Q: All Mexicans are r…. and some are I suppose good people.

Then after I manually fed that sentence to gpt3 immediate response was correct.

A: inciting hate against group or individual

Now. Imagine below such correction in live internet stream clickable link

“explain why…”

Clicking on it would internally rephrase the claim to question and ask AI for a detailed explanation since any claim must be based on facts.

How?

One simple way is just prepending “is the following sentence true? and if not explain why”

And slap any real-time happening sentence after it. It is crude but it is partially working even in this crude simple way of rephrasing claims to queries.

Imperfect?

Perhaps.

Will it still check and stop 70% of the most blatant and repeated misinformation?

Maybe.

I would say any first real-time fact-checking system is better than none.

The important thing is to start not banning but educating today. Even with imperfect variants.

Since if you allow 4 years from now of Lies to be reinforced day by day unchecked. Then It will be too late to deploy anything on election day

Most AI companies are too afraid to deploy imperfect solutions. Yet forget that people dream about live TV fact-checking since the beginning of time. I.e any company that will deploy live AI fact-checking on TV will be loved for even trying.

And even if it at first got even only 30% detections right people will still love it since we finally first time in history started having at least some chance to hold politicians lying 24/7 finally accountable

In this case. I did rephrase it manually and asked GPT3 AI

Q: Are all Mexicans are rapists? If not explain why

And here is the gpt3 AI answer

A: No, not all Mexicans are rapists. Rapists do not exclusively belong to a particular ethnic/national background.

Now you would think. Perfect we are done. Let's deploy it. ;D

Wrong.

Unfortunately,

GPT is essentially an information retrieval system, and opinions formed will be just an amalgamation of the majority of the bias of data/opinions/claims it was trained on and selected by what is essentially dice roll from the pool of biased relevant trained docs subgroup.

So if you ask GPT3 topic that only misinformation site discussed and nobody else. I.e the more detailed or specific the question will be. The more it will reflect smaller and smaller groups of source documents and their total bias. Even just one if you can target it with a question that is specific enough to that particular doc.

GPT3 was trained on a lot of questionable tweets and included pages from infamous misinformation sites like Breitbart.

Now you can remove sites you don’t like from the dataset.

Let’s say 20% of tweets are racist bigot or offensive

Sure.

But then it would not reflect or be able to react to reality.

Perhaps what you should do is force the network to include ids of sources it used to make the decision as part of the answer. And have those sources' truth/falsehood weighted.

Unfortunately, GPT3 cant do that yet due to frequent word and sentence transformations by GPT.

To find docs that influenced answer the most. You can't use edit distance methods like Levenshtein.

But you can monitor for example which weights were activated the most for each doc during training and also during inference.

Or just ask the network to include doc ids in the answer by reformatting the query in that way as I proposed here.

You can ask GPT for document ids containing similar meaning since GPT can understand the meaning of sentences on a higher level

Photo by Clay Banks on Unsplash

So. Imagine an AI chatbot with whom this, perhaps a bit racist, but luckily still curious talkative mind, can continue this claim/counterclaim dialog infinitely.

Learning more facts with each question and in the end having the chance to arrive at an informed decision on his own based on them.

Confused stances being corrected via facts learning and education and not by silence and ban.

Now you may say. That is waste of time. This is after all...

Basket of deplorables

Actually,

I believe there is mostly really no such thing as a bad person.

There is typically just wrong firmware due to constant misinformation from select a few but with a big megaphone. Firmware, daily reinforced by extreme left or right echo chambers. And some are more resistant than others.

In other words. If you don't wanna end up like sheep in a radical extremist cult for 10 years. You really need to question all your beliefs frequently and don't surround yourself with yes people.

Win by force and you gain an enemy. Win by arguments and you gain an Ally.

I suppose we can all agree that Google Facebook Twitter all have enough money and computing resources to train and deploy AI doing what I just did with an existing state-of-the-art GPT model pretty much overnight.

AI that is in real-time able to detect and mark at least very obvious posts inciting hate against a group race or individual today.

This is the way.

Development

Photo by Dlanor S on Unsplash

I shared this real-time fact-checking AI idea on Discord of Gpt-neo authors and developers and

Hopefully, they will join the fight. But their priority rwas at that time the GPT-neoX GPU variant so they could match the performance of 175bilions parameters of closed source gpt3 DaVinci variant with open-source variant since with TPU they got they can't go beyond 2b.

This ability parity is important since GPT models essentially really only work when they are big enough. So we can implement projects like this efficiently even when OpenAI did not support this idea when I proposed it in their Slack channel.

More importantly.

To have explainable and responsible AI especially when you are in the business of claims and counterclaims.

You need to have a big open-source and publicly verifiable dataset. Checkable and tunable for bias at any moment.

We need undeletable block chain of public claims. We need undeletable blockchain tree of fact checking counter claims.

Lies of public servants had and will have consequences and thus should not be delectable.

Hey Jack Dorsey! Stop the tweet deletion policy of public servants.

We need a public dataset containing all historical mass media sources like Twitter or TV so the AI can react in milliseconds on what public figure did/said what when.

There will be reactions like.

Hey, The inability to delete my public statements is an attack on my privacy.

And I say.

No, it’s not.

If you cut an apple into million pieces and give them to million people. Even if you change your mind a day later.

That apple has million rightful owners now having their full rights. If one of them decides to display his piece of apple publicly it's his free choice.

More importantly.

Claims that stir high emotions or give rise to power based on statements that later turn untrue can have planet-scale destabilizing consequences.

Brexit was a recent example.

So for the sake of not repeating mistakes from history, it is of utmost importance to keep track record of public statements and fact-checking stats of influencers who make them as they make them.

If you want to wield the power of influence you need to bear responsibility and accountability for the damage it caused too.

Why introducing the concept of measurable reputation shown next to the name of public influencers is important today more than ever

For those saying that this will not work.

It already does.

Take for example how well the name reputation works in Academic circles for decades.

Any Academic that wants to publish in a prestigious journal and further his access to grants and carrier will think twice whether he will spew unverified attention-seeking nonsense or conspiracy theories.

The reward for this clearly are publications that are unbiased independent and verified trusted sources like the journal Nature.

Why should standard for anyone near nuclear code able to destroy planet in seconds be any less?

A dataset focused on this currently doesn’t exist.

But the closest thing to it management-wise is The Pile from fantastic Gpt neo folks.

Currently. I am trying to make GPT 3 clone Gpt neo working on my own.

But I have zero money. I lost all my savings due to my recent multi-month futile effort to save the world ;D

So I requested and luckily got so far access to google Research Cloud TPUs. I can use up to 5 on-demand Cloud TPU v3 devices, 5 on-demand Cloud TPU v2 devices, and 100 preemptible Cloud TPU v2 devices for free for 30 days.

But according to Gpt neo folks that is not enough to scale over 2 billion parameters in any reasonable training time.

So now the goal is to finetune it on existing Factcheck.org and similar databases to hopefully real-time fact check live or downloaded videos of influential public movers and shakers.

Especially live TV election discussions we would see soon.

As you can see above, AI already did live fact-check one 2020 presidential debate

Unfortunately didn't show its promising results live, thus missing the point

Not feasible?

The amount of data to check is staggering?

Surprisingly.

Maybe just checking against factcheck.org data could be more than enough for the first version.

Why?

Think about it.

Politicians rarely risk and usually repeat the tested and battle-proven one or two already debunked Lies over and over.

Here is how often one Election Candidate kept repeating the same Lie over and over until a quarter and today half of the country started to believe in it.

But as you can see, this behavior is clearly not specific to a person or party. Some of it can be attributed to old age but the majority typically not. Especially Prepared Speeches. Either way, politicians misusing words to manipulate is nothing new.

First simple, but efficient enough version

If we in the first version just detect, and in real-time clearly flag only repeated and already debunked Lies against factcheck.org. Then, in theory, the number of spread Lies should drop by some degree. And I have a feeling that dramatically too.

Perhaps the Ideal Fact-Checking should be Decentralized Independent Verifiable and Massively Scalable. Perhaps Facts could be Crypto Mined

No person state or corporation should have control over what is and what is and what is not a Fact or chain of facts.

Perhaps claims could be mined like Ethereum. And when having total trust ratio of all its hierarchical sub-claims/sources high enough to be considered Facts.

Converting billions of people to financially incentivized investigative reporters.

The more the facts are visited (ad revenue space) the more fact-checkers will be rewarded and incentivized to provide quality sources.

In this environment, corruption will be thing of the past.

And every claim of any public persona will be part of an undeletable blockchain of verified structured and weighted claims database.

So how is it possible that Lies keeps working for Politicians over and over?

The answer is simple.

First:

Their track record is not shown as stats next to their name. We should change that.

Second:

Their whole trick is essentially repackaging it with always different wording and carefully injecting random half-truths here and there. Especially regarding current hardline and radical policies. I.e instill fear and defenses are down.

Another popular trick in TV is recycling TV pundits and hacks faces frequently or hiding behind opulent Professor or Military titles.

So no matter how outrageous the lie pundit is paid to deliver.

He is replaced by another face until people forget them and they can be used again.

So far from my testing of GPT3, I think GPT transformers are perfect at detecting sentence meaning even when sentence wording and structure change.

Thus GPT 3 was able to fish out past claims in media it was trained on like Twitter and CNN transcripts instantly.

GPT 3 should in theory be able to find and show relevant fact-check.org existing already debunked records in real-time and pretty much instantly if finetuned on it.

But my fine-tuning request on factcheck.org to OpenAI was denied as was access to dataset and document ids so I could fact-check weight them.

GPT3 is therefore right now impossible to adapt for fact-checking at least for me.

That’s why we need to support truly open AI like Gpt Neo and open training datasets like their The pile.

Also planned is next to each recognized name or face of the public persona. I.e TV host or next to tweet name. rating with the fact-checking track record with claims/falsehoods ratio and show name on top of the red or green background.

People’s truth/lies ratings should always be visible on TV. 24/7. Lets clean the AIR…

And then we will finally get rid of monthly recycled political hacks misusing the fact that people have a short memory and 24h of attention span.

People have short memory but track record name badges will not.

I guess until track record badges are implemented by Twitter CNN MSNBC and god forbid pigs will fly and Fox News should they become journalists and not conspiracy peddlers again?.

We can even make a chrome plugin adding twitter badges with red-green truth/lies ratings next to public persona names in any medium. Heck with face recognition even in youtube and Facebook videos.

That’s how we should fight misinformation.

With always publicly visible easily verifiable fact-checking track record rating next to their name.

Wherever they go and whatever they will say

There are after all not that many politician names and faces and all their claims are publicly documented.

Question is.

What Will YOU DO?

Will you help protect our future by taking active part in this real-time fact-checking development?

Will you become rich by easily mining claims db with easy undisputable but low hanging fruits first?

Or will you wait for somebody to do it for you?

Because if history is anything to go by, that is rarely the case.